Hi, The mfp package suggests fractional polynomial transformations for non-linear predictors. For continuous variables that take negative values the suggested transformations tend to look ugly, e.g. if x1 is the original variable it is transformed to something like: (x1+5.67)^1 * log(x1+5.67), where 5.67 is the shift to make it positive. Can I somehow modify the mfp function to exclude some powers among those tried (e.g. those involving log and 0.5). I am willing to sacrifice some precision for a more interpretable transformation. Or is there another package to do something similar? Any suggestions very much appreciated! Eleni Medical Statistician Epidemiology & Public Health University College London, UK